An Optimization Approach to Study the Distribution of the Coefficient of Determination, R2.
Abstract
The empirical model builder utilizing regression techniques frequently relies on the coefficient of determination, R sub 2 , to measure goodness of fit. Costing and pricing analysts using such variable selection techniques frequently encounter inflated R sub 2 values. This paper examines the space within which the regression model operates and presents practical optimization algorithms to help assess the amount of confidence that can be placed in R sub 2 for a particular set of candidate predictor variables. The algorithms describe a technique using linear programming to find the lowest value of R sub 2 possible using the given set of data. (Author)
Document Details
- Document Type
- Technical Report
- Publication Date
- Oct 01, 1982
- Accession Number
- ADA124586
Entities
People
- Jeffery Donald Vick
Organizations
- Naval Postgraduate School